Skip to content

Customer data vs machine builder data

github-actions[bot] edited this page Oct 20, 2026 · 1 revision

Customer data vs. machine-builder data

This is the core lesson of Cromley Hydraulics. Two parties look at the same turbine, from the same broker, and need very different data.

Two trees side by side

Customer tree CH tree
Owner DRWPA, the dam owner Cromley Hydraulics, the machine builder
Root v1.0/DRWPA/CopperCanyon v1.0/CH/Fleet/DRWPA/CopperCanyon/<AssetSerial>
Organised by Where things are in the plant (ISA-95: area, module) Which asset it is (serial number), then subsystem and component
Example .../Powerhouse/Unit1/MW .../CH-FR-1987-0042/Bearings/TGB/PadTemp03
Published by The dam simulator (unit controllers, plant controller) One CH PLC per turbine
Used by The dam SCADA, the grid operator CH's service engineers, edge devices, analytics
Size 82 measured topics for the whole dam, plus 27 command topics 74 process topics per turbine, 8 waveform channels, alarms, nameplate, status
Rate 1 s 1 s process values; waveform blocks every 30 s (accelerometer every 5 min) and on alarm
Payload {"value", "unit", "timestamp"} (the Sweet Harmony format) the same plus "quality" and "seq"
Questions it answers Is the plant producing what the grid asked? Is anything tripping? Is the dam safe? How healthy is my machine? What will fail, when? What should the next service visit do?

Why by serial number? A machine builder thinks in assets: the serial CH-FR-1987-0042 stays the same when the customer renames "Unit 1" or the unit is refurbished. The customer's name only lives in _meta/Nameplate (customer.unitName, customer.unitTag).

Unit (customer) Serial (CH) Commissioned
Unit1 CH-FR-1987-0042 1987
Unit2 CH-FR-1987-0043 1987
Unit3 CH-FR-1989-0051 1989

Context: subscribe only

A bearing temperature means little without the operating point. Each CH PLC subscribes a few customer topics and republishes them under Context/, with its own quality and seq:

CH Context/… Copied from the customer tree
MW, MVAr Powerhouse/<unit>/MW, MVAr
GatePosition_pct Powerhouse/<unit>/GatePosition
NetHead_m Powerhouse/<unit>/NetHead
Speed_rpm Powerhouse/<unit>/Speed
UnitState Powerhouse/<unit>/State
TailwaterLevel_m River/TailwaterLevel

Also Governor/HPU/GateSetpoint comes from Powerhouse/<unit>/GateSetpoint. A context value older than 5 s becomes UNCERTAIN (last known value), never received BAD. The CH PLC never writes into the customer tree.

The overlap: dual-element sensors

A few sensors are dual-element: element A goes to the customer's unit controller (protection), element B to the CH PLC. The customer gets one RTD per bearing (pad 1), one cooling water flow meter and the governor oil pressure; CH gets every pad, every circuit and much more. See the table on Architecture.

That is why the two parties notice faults at different times:

Scenario Customer sees CH sees
A. TGB pad 3 degrading Its pad-1 RTD rises ~3 K: nothing alarms Pad 3 climbs from ~64 to ~95 °C; pad-to-pad spread grows; PLC alarms at 75 and 85 °C
B. Cooling-water strainer Cooling water flow low within minutes; bearing alarms later; a trip if nobody acts The same, per circuit, plus the oil cooler outlet and every pad
F. Unbalance Nothing: the customer has no vibration probes 1X amplitude ~4×, phase shift, PLC vibration alert
G. CH RTD drift Nothing: element A is fine One pad drifts +20 °C while its neighbours don't, quality still GOOD

Payloads, retained messages, last will

  • Customer: {"value": 36.65, "unit": "MW", "timestamp": "2026-10-20T12:00:00.000Z"}, retained.
  • CH: {"value": 61.4, "unit": "°C", "timestamp": "...", "quality": "GOOD", "seq": 128734}, retained by default (RETAIN_PROCESS_VALUES). seq counts every message one PLC sends, so a subscriber can detect gaps. Waveforms and alarm events are not retained.
  • Last will: each CH PLC's MQTT session sets _meta/Status to OFFLINE if it disappears, and publishes ONLINE (retained) when it connects. The customer tree has no such status: the SCADA detects stale data by age.

Full formats on the Namespace page.

Who owns the data?

Notes for discussion, not legal advice:

  • Under the EU Data Act, users of connected products have data access rights since 12 September 2025, and products placed on the market after 12 September 2026 must offer access by design.
  • CH may only use non-personal product data under an agreement with the customer.
  • Selling data to third parties needs the customer's consent; anonymised fleet benchmarks are more realistic than raw data.

The question for vendors

Everything is on one broker, with no security. A real site must keep customer control data and machine-builder monitoring apart, and send off site only what was agreed. Each edge vendor answers:

  • How do you scope what leaves the site to the CH tree (or an agreed subset)?
  • How do you guarantee nothing is ever written into the customer tree?
  • How do you respect the customer's consent when it changes?

See the Edge vendor brief and Business model.

Clone this wiki locally